Node Requirements
What a node needs before it runs. Required inputs, a named connection when the catalog says so, and a model on AI nodes.
Node requirements
Each catalog node declares inputs, outputs, and whether it needs a credential. The inspector is the source of truth for field names. This page is the shared pattern, not a second catalog.
Start nodes
A flow needs one start node:
| Node | Slug | Role |
|---|---|---|
| Start | start | On-demand, API, webhook |
| Start Scheduled | startScheduled | Recurring, weekly, or runAt |
| Start Subflow | startSubflow | Invoked by a parent Subflow node |
There is no Email, API, Webhook, or External Event start node. See Manual.
Start nodes do not accept upstream data handles. Downstream nodes read trigger payload from the start node (body, headers, query, method on webhook/API) or from Flow inputs.
Required inputs
A required input must be one of:
- A static value in the inspector
- A mapped expression / handle from an upstream output
- Left empty, then supplied at run time as
{nodeId}_{inputName}
If it is still empty at enqueue, validateFlow fails for connected nodes.
Generate Text requires Prompt (prompt, data type prompt) and a selected model. Temperature is an advanced slider from 0.0 to 1.0, not 0-2.
Credentials
If the node uses an integration, pick a named organization connection. See Credential management.
HITL email is split: platform Resend vs named Resend key vs named SMTP. See HITL Approvals.
Unconnected nodes
Nodes not reachable from the start node are not validated and do not run. They can sit on the canvas while you build.
Outputs
Outputs are declared on the catalog row. You do not have to wire every output. Cost and token fields on AI nodes are run outputs when the platform records them. They are not a reason to invent a cost handle in docs if the inspector does not show it.